English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Deep reinforcement learning (DRL) is a promising approach for developinglegged locomotion skills. However, the iterative design process that isinevitable in practice is poorly supported by the default methodology. It isdifficult to predict the outcomes of changes made to the reward functions,policy architectures, and the set of tasks being trained on. In this paper, wepropose a practical method that allows the reward function to be fullyredefined on each successive design iteration while limiting the deviation fromthe previous iteration. We characterize policies via sets of DeterministicAction Stochastic State (DASS) tuples, which represent the deterministic policystate-action pairs as sampled from the states visited by the trained stochasticpolicy. New policies are trained using a policy gradient algorithm which thenmixes RL-based policy gradients with gradient updates defined by the DASStuples. The tuples also allow for robust policy distillation to new networkarchitectures. We demonstrate the effectiveness of this iterative-designapproach on the bipedal robot Cassie, achieving stable walking with differentgait styles at various speeds. We demonstrate the successful transfer ofpolicies learned in simulation to the physical robot without any dynamicsrandomization, and that variable-speed walking policies for the physical robotcan be represented by a small dataset of 5-10k tuples.

Iterative Reinforcement Learning Based Design of Dynamic Locomotion Skills for Cassie